Upload
xdboy2006
View
223
Download
0
Embed Size (px)
Citation preview
7/28/2019 Zaspil43 Morimoto
1/27
Agreement Properties and Word Order in
Comparative Bantu
Yukiko Morimoto
ZAS, Berlin
Agreement is traditionally viewed as a cross-referencing device for core arguments
such as subjects and (primary) objects.1 In this paper, I discuss data from Bantu lan-
guages that lead to a radical departure from this generally accepted position: agree-
ment in a subset of Bantu languages cross-references a (sentential) topic rather than
the subject. The crucial evidence for topic agreement comes from a construction
known as subject-object (S-O) reversal, where the fronted patient agrees with what
has uniformly been taken to be a subject marker. The correct analysis of S-O re-
versal as a topic construction with topic agreement explains a range of known facts
in the languages in question. Furthermore, synchronic variation across Bantu in the
presence/absence of S-O reversal and in the properties of the (topic/subject) agree-
ment marker suggests a diachronic path from topic to subject marking. The sys-
tematic variation and covariation in the syntax of Bantu languages and the historical
picture that it offers would be missed altogether if we continue to reject the idea thatthe notion of topic can be deeply grammaticized in the form of agreement.
1 Introduction
One of the most pronounced and most studied aspects of Bantu grammar is
the rich classes of agreement morphology. One of the most detailed studies ofBantu agreement within generative syntax is offered by Bresnan and Mchombo
(1987), who focus on the properties of subject and object marking in the Eastern
Bantu language Chiche wa. They show that in Chiche wa the subject marker (SM)
functions both as grammatical agreement and as a topic-anaphoric pronoun,
whereas the object marker (OM) functions only as a promoninal argument. Thus
1I am grateful to Peter Sells for discussions on earlier versions of this work, to Alxandre
Kimenyi and Jeanine Ntihirageza for assistance with Kinyarwanda data and Juvenal Ndayiragije
for Kirundi data. I also benefitted from useful comment and suggestions from the editors of
this volume, although unfortunately some issues the editors have raised have to be left for
future work. Thanks also to Christian Geng for editorial assistance. Any remaining errors or
misrepresentations are my own.
161
7/28/2019 Zaspil43 Morimoto
2/27
Yukiko Morimoto
in the absence of the OM, the object NP must be licensed positionally within VP.
They also observe that the facts about object marking in Chichewa do not auto-
matically extend across the whole Bantu family. For example, Swahili requires
object marking for definite objects (cf. Bokamba 1981; also Wald 1979); in the
Imithupi dialect of Makua studied by Stucky (1981, 1983), the OM is obliga-
tory for the human classes (classes 1 & 2) even when the overt object NP is
not topical. Morimoto (2002) discusses these facts under the general concept
of differential object marking (DOM; cf. Bossong 1985) and suggests that these
cross-linguistic data can be viewed as various stages of the gradual loss of the
inflectional morphology and rise of positional licensing for objects.
A closer look at subject marking in other Bantu languages also reveals
that facts in those languages diverge from the findings reported by Bresnan and
Mchombo on Chiche wa. Bresnan and Mchombo (p.778) note that in Dzamba,
although all arguments inside VP can be questioned in situ, as in (1b), the sub-
ject cannot be questioned in its initial position, as shown by the ungrammatical-
ity of (1c). In order to question the subject, a relative clause must be used, as
shown in (1d).2
(1) a. o-Nebo a-imol-aki o-Biko e-kondo loo me. Dzamba
Nebo told Biko a story/tale today.
b. o-Nebo a-imol-aki nzany e-kondo loo me?(lit.) Nebo told who a story today?
c. *nzany a-imol-aki o-Biko e-kondo loo me?
Who told Biko a story today?
d. o-Moto o-wimol-aki o-Biko e-kondo loo me nzany?
(lit.) The person who told Biko a story today is who?
Based on such data, Bresnan and Mchombo speculate that unlike Chichewa, in
which the subject can be questioned in its initial position, in Dzamba, all sub-
ject NPs must necessarily be grammaticized topics. They further note, citing
Bokamba (1981), that Dzamba has nominal pre-prefixes which are obligatory
on definite NPs. For example, in (1), we have the pre-prefix o- on the proper
nouns (o-Nebo, o-Biko). Such pre-prefixes are obligatory on subjects. This
supports their hypothesis that all subjects are obligatorily grammaticized top-
2In the examples taken from published sources, I have kept the original use of tones and
glosses, which, in some cases, results in inconsistency throughout this paper. For example, a
final vowel (FV) in Chichewa (e.g. Bresnan and Mchombo 1987) is glossed as ASP(ECT) in
Kinyarwanda (e.g. Kimenyi 1980); the morpheme ra is glossed as anti-focus (AF) in Kirundi
(Ndayiragije 1999) but present tense (PRES) in Kinyarwanda (e.g. Kimenyi 1980).
162
7/28/2019 Zaspil43 Morimoto
3/27
Agreement Properties and Word Order in Comparative Bantu
ics. Taking this initial observation as a starting point, in this paper I explore the
idea that what has uniformly been taken to be the subject marker in Bantu lan-
guages in fact represents two types of agreement system: subject versus topic
agreement.
The crucial evidence for topic agreement in Bantu languages comes from
a construction known as subject-object (S-O) reversal in a subset of Bantu lan-
guages, exemplified in (2) from Kinyarwanda (Kimenyi 1980).3 In the canon-
ical SVO sentence in (2a), the relevant agreement marker a (boldfaced) agrees
with class 1 of the logical subject Umuhuungu boy. In the reversal sentence
in (2b), the agreement marker ki agrees in class 7 of the lower argument igitabo
book. As shown in the translation, the fronted patient (book) is a topic, and
the postverbal logical subject receives a (contrastive) focus interpretation.4
(2) a. Umuhuungu a-ra-som-a igitabo. SVO
1boy 1-PRES-read-AS P 7book
The boy is reading the book.
b. Igitabo ki-som-a umuhuungu. OVS
7book 7-read-AS P 1boy
The boy (FO C) is reading the book (TO P) Kinyarwanda
The agreement pattern we see in (2b) is the unique property of this con-
struction that raises the question as to the grammatical relation of the fronted
patient to the verb. Unlike the more familiar left-dislocation, the object marker
agreeing with the fronted patient is never present in S-O reversal, as shown the
contrast between S-O reversal (3a) and left-dislocation (3b) (Kimenyi 1980:192,
ex.(4)).5
3The data without citation have been generously provided by Alexandre Kimenyi for Kin-
yarwanda and Juvenal Ndayiragije for Kirundi. Tones are unfortunately left out in the elicited
examples.4The abbreviations used in the glosses are as follows: AF=anti-focus, ASC=associative,
ASP=aspect, Dz=Dzamba, ENCL=enclitic, exp=experiencer, FV=final vowel, IM-
PERF=imperfect, IM.FUT=immediate future, INDIC=indicative, INF=infinitive (=class
15), Krw=Kinyarwanda, Kdi=Kirundi, LOC=locative, number N= noun class, OM=object
marker, PASS=passive, PERF=perfect, PRES=present tense, PST=past, SM=subject marker
TM=topic marker.5The S-O reversal example in (3a) is constructed by the author.
163
7/28/2019 Zaspil43 Morimoto
4/27
Yukiko Morimoto
(3) a. Igitabo ki-som-ye umwaana. S-O reversal
7book 7-read-AS P 1child
The child (FO C) is reading the book (TO P)
b. igitabo, umwaana a-ra-gi-som-ye. Object left-dislocation
7book 1child 1-PRES-7-read-AS P
The book, the child is reading it. Kinyarwanda
There are two possible solutions for the agreement pattern. The predominant
solution has been to maintain the standard assumption about agreement that it
licenses core argument functions such as subjects and (primary) objects. Under
this assumption, the agreement marker in question must be a subject marker; S-Oreversal such as (2b) is then a grammatical-relation changing operation whereby
the patient gets linked to the grammatical subject, and the logical subject to the
grammatical object. I refer to this solution as the subject analysis. An alternative
solution, referred to as the topic analysis, assumes that no grammatical relation
change takes place in S-O reversal. Rather, the agreement marker (ki in (2b)) is
analyzed as a topic marker licensing the topical object. These two alternative
solutions naturally lead to diverging predictions. I show evidence to support the
topic analysis and argue that the topic function is indeed among the functionsthat can trigger agreement in so-called topic-prominent languages like Bantu.6
The discussion in the rest of the paper proceeds as follows. In section 2, I
review some arguments for the previously proposed subject analysis, and point
out why the analysis, while being able to maintain the traditionally accepted
view of agreement, is nonetheless untenable for empirical reasons. In section 3
I provide the topic analysis within the constraint-based framework of Lexical-
Functional Grammar (LFG; Bresnan 1982, 2001, Dalrymple, Kaplan, Maxwell,
& Zaenen 1995). In section 4 I try to relate the data on S-O reversal and othercorrelating facts to the historical path these languages apparently have under-
gone with respect to their agreement system. The final section summerizes the
discussion and points out remaining issues for further research.
6The topic analysis of S-O reversal argued in the present work stems from a discussion I had
with Joan Bresnan (July 1999) in relation to the observation that in a subset of Bantu languages
such as Dzamba in (1), the subject is always a grammaticized topic (noted in Bresnan and
Mchombo 1987). I benefitted particularly from our discussion on the systematic differences
between the two similar constructionslocative inversion (Bresnan and Kanerva 1989, Bresnan
1994) and S-O reversal.
164
7/28/2019 Zaspil43 Morimoto
5/27
Agreement Properties and Word Order in Comparative Bantu
2 Against the Subject Analysis of Reversal
Earlier work on S-O reversal (e.g. Bokamba 1976, 1979, 1985, Kimenyi 1980,
1988, Morimoto 1999) has generally assumed that the construction involvesgrammatical relation change, mainly due to the fact that the preverbal patient
in reversal apparently triggers subject agreement: the patient (object) is fronted
to the subject position and assumes the subject function, and the agent (logical
subject) is postposed to the VP-internal object position. The grammatical func-
tion (GF) reversal analysis is somewhat analogous to the analysis of inversion
in other languages: English, Chiche wa, and Chishona, for example, exhibit in-
version of the theme subject and locative complement (e.g. Out of the bushes
appeared a giant bear; cf. Bresnan and Kanerva 1989, Harford 1990, Bresnan1994); Sesotho and Setswana additionally allow inversion that involves (intran-
sitive) unergative predicates (e.g. equivalent ofIn the field are grazing the cattle;
see Machobane 1995, Demuth and Mmusi 1997, Ldrup 1999). In these con-
structions, the preverbal locative is analyzed as the grammatical subject and the
logical subject in postverbal position as object, inducing a grammatical relation
change; see Morimoto (1999) for an earlier OT account of all three language
types including S-O (agent-patient) reversal in Kirundi and Kinyarwanda. The
postverbal agent in S-O reversal has also been analyzed not as an object but as
a demoted argument (e.g. chomeur in Kimenyis (1980) analysis in Relational
Grammar). Studies within the Principles and Parameters approach have ana-
lyzed S-O reversal as involving raising of the object to some functional projec-
tion usually reserved for the subject (e.g. SpecIP in Kinyalolo (1991), SpecTP
in Ura (1996), Ndayiragije (1999)) and verb raising (to I or T). The logical sub-
ject either remains in its VP-internal position (cf. Kinyalolo, Ura) or raises to a
specifier of FocusP above VP (cf. Ndayiragije). The latter (FocusP) alternative
is intended to capture the focalization effect of this construction.
Apart from agreement, however, little evidence has been provided for the
grammatical function status of the reversal arguments. Kimenyi (1980), for ex-
ample, proposes that S-O reversal is a type of subjectivization process akin to
passive: the non-agent argument canonically realized as a non-subject is real-
ized as subject, and the logical subject is demoted. The demotion analysis is
supported by the observations that the postverbal agent cannot be relativized,
passivized, clefted or pronominalized. These are taken to be standard diag-
nostics for termhood within Relational Grammar adopted in Kimenyis work.
Kimenyi, however, admits that the preverbal NP lacks the properties of the usualsubjects apart from agreement: NPs advanced to subject by the [S-O] reversal
rule do not acquire the properties of basic subjects, such as raising, deletion
165
7/28/2019 Zaspil43 Morimoto
6/27
Yukiko Morimoto
under identity, and ha-insertion; the only subject property they acquire is verb
agreement (Kimenyi 1980:145).
Despite the apparent formal similarities between locative inversion and
S-O reversal and attempts at a unified analysis of these phenomena, I show thatthere are several advantages for not adopting the subject analysis, and that the
alternative analysis pursued in the present work which crucially relies on the
grammaticized notion ofTOPIC enables us to explain a broader range of related
facts. The correct analysis of S-O reversal is critical to the understanding of the
agreement systems in Bantu. Although in my analysis I assume that there is no
actual GF reversal, I will continue to refer to this construction as S-O reversal
for descriptive purposes.
2.1 Lack of Evidence for Grammatical Relation Change
In the studies of locative inversion in Chiche wa and English, Bresnan and Kan-
erva (1989) and Bresnan (1994) apply a number of syntactic tests for sub-
jecthood and show conclusively that preverbal the inverted locative is indeed
the grammatical subject, supporting their view that locative inversion involves
grammatical relation change. Here I briefly consider VP attribution, gappingin coordinate structure, and subject-to-subject raising. I then comment on each
test with respect to Kirundi and Kinyarwanda, for which elicited (analogous but
not identical or even comparable) data are available, only to point out that these
tests cannot be reliably used to show anything about the grammatical status of
the preverbal patient in S-O reversal.
Control of Attributive VPs: Bresnan (1994:93 ex (58)) shows that
Chichewa has a non-finite verb that can be used to modify NPs, as in (4).
(4) a. m-sodzi [V P w-o-k-a nsomba pa m-pando] Chichewa1-fisherman 1-AS C.INF-put-FV 10.fish 16 3-chair
a fisherman putting fish on the chair
b. nsomba [V P z-o-k--dw-a pa m-pando]10.fish 10-AS C.IN F-put-PASS-FV 16 3-chair
fish being put on a chair
In both active and passive examples in (4), the subject is modified by the infini-tival VP. In Chichewa locative inversion, Bresnan (p.94, ex (59)) shows that the
inverted locative subject can also be modified by a non-finite VP:
166
7/28/2019 Zaspil43 Morimoto
7/27
Agreement Properties and Word Order in Comparative Bantu
(5) m-nkhalango m-o-khaal-a mi-kango Chiche wa
18-9.forest 18-AS C.INF-live-FV 4-lion
in the forest where there live lions
The example in (5) thus provides evidence for the subject status of the inverted
locative in Chichewa.
Extraction out of Coordinate Constituents: Extraction out of coordi-
nate constituents is used to distinguish among grammatical functions. The gen-
eralization that holds for English is that subject gaps at the top level of one
coordinate constituent cannot occur with any other kind of gap the other coor-
dinate constituent, as succinctly stated in Bresnan 1994:98 (cf. Williams 1977,
Gazdar 1981, Falk 1983, Woolford 1987). This is illustrated by the examples in(6) (Bresnan 1994:98, exs (71)(72)).
(6) a. Shes someone that loves cooking and hates jogging. (S-S)
b. Shes someone that cooking amuses and jogging bores . (O-O)
(7) a. *Shes someone that cooking amuses and hates jogging. (O-S)
b. Shes someone that cooking amuses and I expect will hate jogging.
(O-embedded S)
Bresnan (1994) shows how the inverted locative in locative inversion in English
(e.g. Out of the bushes appeared a monster) observes the same constraint with
respect to extraction from coordinate constituents (Bresnan 1994:98, exs (73)
(74)). This test has been used to test for subjecthood in other languages as well
(cf. Joshi 1993 for Marathi, Kroeger 1993 for Tagalog), as this allows us to
distinguish between subjects and non-subjects.
(8) a. Thats the old graveyard, in which is buried a pirate and is likelyto be buried a treasure. (subj-subj)
b. Thats the old graveyard, in which workers are digging and a treasure
is likely to be buried . (nonsubj-nonsubj)
(9) a. ??Thats the old graveyard, in which workers are digging and they say
is likely to be buried a treasure. (nonsubj-subj)
b. Thats the old graveyard, in which workers are digging and they say
is buried a treasure. (obl-embedded subj)
These examples, then, constitute, evidence that the inverted locative is the gram-
167
7/28/2019 Zaspil43 Morimoto
8/27
Yukiko Morimoto
matical subject.
Raising: Bresnan and Kanerva (1989:14) note that Chichewa has very
few subject raising predicates, but nonetheless report one case of raising of the
locative subject of the infinitive:
(10) . . . pa-chi-dzala pa-funa ku-tukumbuka chi-nthu . . .
16-7-rubbish pit 16-IM.FU T-want IN F-emerge 7-thing
. . . there seems to be something coming out of the rubbish pit . . .
Though not reflected in the English translation, the form of the sentence in (10)
is locative inversion like the English From the rubbish pit seems to be emerg-
ing something. Since such raising is universally restricted to the grammatical
subject of the infinitive, the example in (10) confirms the subject status of theinverted locative.
S-O Reversal Languages: As it turns out, these tests for subjecthood that
are successfully applied for the conclusive results in Chichewa and English are
not applicable to the reversal languages, Kirundi and Kinyarwanda.
First, in Kirundi, there is no construction analogous to VP attribution in
Chichewathat shows subject control. The closest construction in Kirundi is a
clausal adjunct that translates to something like once he has finished buying
. . . or after having bought . . . , according to Ndayiragije (p.c. May 2000).7
Second, extraction out of coordinate constituents fails to show anything about
grammatical functions in Bantu languages for obvious reason: whenever there
is a gap (null elements), the subject/object agreement morphology on the verb
functions as topic-anaphoric agreement. Hence there can never be a real gap
in the coordinate structure. Lastly, Kirundi and Kinyarwanda simply have no
subject raising predicates such as seem, expect, and likely.
In short, the standard tests for subjecthood illustrated above for Chichewa
and English all fail to apply to the reversal languages (Kirundi and Kin-
yarwanda) due to the lack of comparable constructions/verb types.
2.2 Unexplained Facts
The subject analysis put forth by various researchers also leaves some facts un-
explained. Here I focus on the properties of the postverbal agent in S-O reversal,
and variation across Bantu languages in the properties of agreement.
7See Morimoto 2000:155 for the relevant data.
168
7/28/2019 Zaspil43 Morimoto
9/27
Agreement Properties and Word Order in Comparative Bantu
The postverbal agent has no object properties
One of the puzzling facts that remain unexplained under the subject analysis is
that the postverbal agent of a reversal sentence does not display usual object
properties. For example, it can neither be relativized (11) nor promoninalized
(12).8
(11) *N-kund-a umuhuungu igitabo ki-som-a. Kinyarwanda
I-like-AS P 1boy 7book 7-read-AS P[Intended as] I like the boy (FO C) who is reading the book (TO P).
[Interpreted as] I like the boy that the book is reading (by magic)
(12) a. Ba-ra-gi-som-a.
2-AF-7-read-AS P
They are reading it.
b. Cyi-(ra)-ba-som-a.
7-(AF)-2-read-AS P#It is reading them. (non-reversal reading)
*They (FO C) are reading it (TO P). (reversal reading)
The postverbal agent is also not in the canonical, immediately postverbal, object
position. In (13a), the postverbal agent is positioned after the adverbial phrasebuhurobuhuro slowly, presumably adjoined to VP. Placing it in the canoni-
cal object position inside VP results in ungrammaticality as illustrated in (13b).
This postverbal agent is therefore neither positionally nor morphologically li-
censed as the grammatical object.
(13) a. Ibitabo bi-a-som-ye buhurobuhuro Yohani. Kirundi
book 7-PAST-read-PERF slowly John
John (FO C) read the book (TO P) slowly.
b. *Ibitabo bi-a-som-ye Yohani buhurobuhuro.
The postverbal logical subject also could not be a demoted oblique as ana-
lyzed by Kimenyi (1980): in the languages considered here (e.g. Kirundi, Kin-
yarwanda, Dzamba), oblique arguments are generally marked by prepositions
8The ungrammaticality of example (11) does not improve even if we add an object marker
corresponding to the relativized head noun umuhuungu boy or the correct relative tone mark-
ing on the verb inside the relative clause, as shown below.
(i) a. *N-kund-a umuhuungu igitabo ki-mu-som-a.b. *N-kund-a umuhuungu igitabo ki-som-a.
Thanks to Jeanine Ntihirageza for this additional information.
169
7/28/2019 Zaspil43 Morimoto
10/27
Yukiko Morimoto
such as n in (14) from Kinyarwanda. If the postverbal agent were a demoted
oblique, this would be an odd exception, for which no explanation has been
provided.9
(14) a. Umwaalmu a-ra-andik-a n karamu. Kinyarwandateacher 1-PRES-write-AS P with pen
The teacher is writing with a pen.
b. Umwaana y-a-rir-aga n agahiinda keenshi.
child 1-PS T-cry-AS P with sorrow much
The child was crying with much sorrow. Kimenyi (1980)
If the postverbal agent is not an object or oblique, then we are left with the
assumption that it is the grammatical subject.
Reversal and non-reversal languages show varying properties of agreement
A closer look at the properties of the agreement morphology in question across
the Bantu family reveals systematic variation that correlates with the pres-
ence/absence of S-O reversal. The relevant data include the patterns of wh-
questions and idiomatic expressions involving the verb and object. These data
both point towards the different properties of the agreement morphology acrossBantu.
Wh-elements: Wh-phrases are inherently focused, given that they ask for
new information. If the reversal languages employ topic rather than subject
agreement and requires the agreeing NP to be always topic, then these lan-
guages would not allow a wh-element to agree. We saw this earlier in (1c) from
Dzamba. Another (similar) example is given in (15). The same holds for other
reversal languages like Kinyarwanda as shown in (16). Chichewa on the other
hand does not permit S-O reversal, and it allows wh-subjects in situ, as shown in(17)).10
(15) *Nzany o-wimol-aki o-Biko e-kondo l me? Dz. wh-subject in situ
Who told Biko a story/tale today? (Bokamba 1981)
9The alternation of the class 1 agreement ay seems to be phonologically conditioned.10Sam Mchombo (p.c, January 2003) points out that the verb in (17) has high tones as in-
dicated, which signal extraction of the wh-element. In other words, this question still appears
to retain the tonal property of the cleft construction (which is built on relativization in Bantu,
according to Bresnan and Mchombo 1987) that is typically used to form a subject wh-question
in Bantu. It could be that the tonal distinction on the verb remains merely as a historical source
of what used to be the cleft required for a subject wh-question.
170
7/28/2019 Zaspil43 Morimoto
11/27
Agreement Properties and Word Order in Comparative Bantu
(16) *Nde y-a-som-ye igitabo?
who SM -PAST-read-ASP book
Who read the book? Krw wh-subject in situ
(17) (Kod) chyani chi-na-onek-a?
Q what(7) SM(7)-PAST-happen-INDIC
What happened? Chichewa wh-subject in situ
If we uniformly assume subject agreement across the whole family, then these
facts would be left unexplained. If we assume, instead, that reversal languages
display topic agreement, the ungrammaticality of (15)(16) would be an obvi-ous consequence of the system.
Verb-object idioms: In idiomatic expressions consisting of a verb and its
object such as (18) in German, the object is non-referential, and as such they
can never be topic.
(18) a. das Kind mit dem Bade ausschutten
to throw the baby with the bath water
b. Viele Koche verderben den Brei.
(Too) many cooks spoil the meal.
If these idiomatic objects can be passive subjects and still maintain the idiomatic
interpretation, then it would show that the language allows non-topical, non-
referential subjects. Now, German has a non-referential expletive subject es,
as in es regnet it is raining. And we see in (19) that the objects in these
verb-object idioms can be passivized subjects without losing the idiomatic in-terpretation.
(19) a. Das Kind wurde mit dem Bade ausgeschuttet.
The baby got thrown with the bath water.
b. Der Brei wird von vielen Kochen verdorben.
The meal gets spoiled by (too) many cooks.
In Chichewa, we also find verb-object idioms, as exemplified in (20a). Thegrammaticality of (20b) shows that Chichewa allows the objects of verb-object
idioms to be passivized (Bresnan and Mchombo 1987:763).
171
7/28/2019 Zaspil43 Morimoto
12/27
Yukiko Morimoto
(20) a. Chifukwa cha mwano wake Mavuto tsopano
because of rudeness his Mavuto now
a-ku-nongonez-a bondo.
SM -PRES-whisper.to-INDIC kneeBecause of his rudeness, Mavuto is now whispering to his knee (= feel-
ing remorse).
b. Bondo li-na-nongonez-edw-a.
knee(5) SM(5)-PAST-whisper.to-PASS-INDIC
The knee was whispered to (= the remorse was felt).
Chichewa also has a few proverbs that involve idiomatic subjects, exemplified
in (21) from Bresnan and Mchombo (1987).
(21) a. Kalulu a-na-mu-omba maondo.
Hare SM -PAST-OM-knock kneesThe hare knocked him in the knees.
(= He has stiff knees which cannot bend.)
b. Njovu i-na-ponda-po.
Elephant 9SM -PAST-step-LO C.ENCL
The elephant stepped on it (hence rendered it dysfunctional). Chichewa
Again in such expressions, the subject is non-referential; hence it cannot be
topical. These data therefore show that subjects in Chichewa need not be topical
or even referential. This is not true of reversal languages, as illustrated in (22)
from Kirundi. The data in (22) are consistent with those of another reversal
language, Dzamba, which requires the agreeing preverbal NP to be topic as we
have already seen in (15).
(22) a. Yohani a-a-ra-hend-ye umunwa.John 1SM -PAST-AF-cheat/mislead-PERF mouth(lit.) John cheated/misled the mouth.
(= John ate almost nothing.)
b. *Umunwa u-a-ra-hend-u-ye (na Yohani).
mouth SM -PAST-AF-cheat/mislead-PASS-PERF by John(lit.) The mouth was cheated/misled (by John).
(= Almost nothing was eaten (by John).) Kdi
In short, these data on idiomatic expressions also reveal different agreement
properties that are unexplained by the uniform subject analysis.
172
7/28/2019 Zaspil43 Morimoto
13/27
Agreement Properties and Word Order in Comparative Bantu
Agentive object without verbal morphology
Apart from the above facts that are unaccounted for in the subject analysis,
it is also worth pointing out that it is typologically quite rare for an agent to
be linked to the object function without any morphology accompanying such
unusual linking (also argued by Ldrup 1999). And this would be precisely the
claim made by the subject analysis.
Now in ergative languages, the more patient-like argument (O) is linked
to the most unmarked absolutive case, along with the sole argument of the in-
transitive predicate (S). The agent of the transitive (A) is linked to the marked
ergative case. The linking patterns of A, S, and O are thus said to be the in-
verse of those in nominative-accusative languages (cf. Manning 1996). Simi-
larly in Austronesian languages, the unmarked linking pattern is one in whichthe patient-like argument is linked to nominative (e.g. ang-marking in Tagalog;
cf. Kroeger 1993).
These inverse linking types, however, display rich case/verbal morphol-
ogy. Bantu languages share no typological properties with these language types.
It is therefore not reasonable to assume that ergative/Austronesian type linking
is shared by (a subset of) Bantu languages without any morphology.
To summarize, in this section I hope to have shown that the predomi-nantly held view of S-O reversal represented by the subject analysis is unten-
able, firstly because there is no reliable evidence for the assumed grammatical
relation change, and secondly because it fails to provide a coherent explanation
of the set of covariation observed in the reversal and non-reversal languages.
3 Salience-Based Agreement vs. Role-Based Agreement
In this section I provide a sketch of the alternative topic analysis ofS-O reversaland agreement patterns across Bantu more generally. The key idea is that Bantu
languages exhibit two types of agreement: salience-based and role-based. Fur-
thermore, the observed variation in agreement properties across the synchronic
Bantu grammars suggests a historical path from salience-based to role-based
agreement.
The idea explored in the topic analysis is that just as arguments are ranked
according to the argument hierarchy, they are ranked according to the topicality
hierarchy. This hierarchy is determined by relative salience of the arguments,and is distinguished from discourse prominence. The notions of topic salience
and discourse prominence are defined in (23).
173
7/28/2019 Zaspil43 Morimoto
14/27
Yukiko Morimoto
(23) a. Topic Salience: Arguments are ranked according to topicality. An ar-
gument that is most prominent in discourse is also the most salient in
terms of topicality. In the absence of such an argument, by default the
nominative argument will be most salient (cf. Givon 1976, Keenan 1976,Li and Thompson 1976).
b. Discourse Prominence: Any element that can bear a (lexical) stress can
be discourse-prominent. It may be topical or focal.
Given that the arguments of a given predicate are ranked according to topic
salience, naturally there can be at most one argument that is most salient in
terms of topicality, while there can be multiple constituents that are discourse-
prominent.11
The claim in (23a) that the nominative argument will be topically the most
salient by default is supported by the cross-linguistic characterization of sub-
jects and the historical source of subjects topic prominence. Givon (1976),
for example, notes that subjects in many languages, such as Mandarin, Mala-
gasy, and also Kinyarwanda, must be either definite or generic (see also Keenan
1976, Li and Thompson 1976). This property is reminiscent of (dislocated) top-
ics, which, according to Givon (1976), are universally restricted to be definite or
generic (also see Alsagoff 1992:192, Lambrecht 1994, Birner and Ward 1998).Malagasy is an VOS language, and the nominative NP is placed sentence fi-
nally and must be topical (= topically the most salient element). However, like
most verb-initial languages, the position for discourse-prominent constituents
is clause-initial.
Even in so-called subject-prominent languages such as English, which tol-
erate indefinite, non-referential subjects, subjects are nonetheless overwhelm-
ingly definite and referential (Givon pp.154155). In English, continuing topic
is generally expressed as a subject, but the new topic (discourse-prominent) isexpressed in a left-dislocated position. For example in an answer to the ques-
tion in A below, B would be more natural than B , as shown in (24), where they
(family) is established information:
(24) A: Where does your family live?
B: They live in N.Y.
B: #My family, they live in N.Y.
11By emphasizing lexical stress in (23b), I mean to exclude cliticized unstressed pronouns
that cannot be discourse-prominent, as in the German (e)s in ich kanns machen I can do it or
the English im in I gotim (I got him).
174
7/28/2019 Zaspil43 Morimoto
15/27
Agreement Properties and Word Order in Comparative Bantu
On the other hand, as a continuation of the above exchange, if B wanted to
mention where his/her cousin lives as opposed to his family, then an utterance
like that of B in (24) becomes most felicitous. In this context, cousin in (25) is
introduced as a new topic contrasting with the old topic family.
(25) B: (Now) my cousin, he lives in Chicago.
B: #(Now) he lives in Chicago, my cousin.
B: (Now) my COUSIN lives in Chicago.
(OK with appropriate intonation)
Thus, a dislocated topic is used as a new/contrastive topic, while the old topic
tends to be expressed as a clause-internal topic or the subject in a non-dislocated
subject construction. Birner and Ward (1998) make a similar pragmatic differ-ence between English dislocation and topicalization based on a large number
of corpus data. I assume therefore that topic salience is universally part of the
argument linking domain. Following the system of abstract case features fa-
miliar in Lexical Decomposition Grammar (e.g. Wunderlich 2000), I represent
topic salience by a binary feature [ht], as shown in (26a,b). The feature [ht]states that there is no argument that is higher in topic salience, and specifies
the element in highest topic salience. Conversely, [+ht] states that there is an
argument that is higher in topic salience. These features form a universal scale
as shown in (26c), which I refer to as the TOPICALITY HIERARCHY.
(26) Encoding topic salience (a la Wunderlich 2000)
a. [ht]: There is no other argument that is higher in topicality.b. [+ht]: There is another argument that is higher in topicality.
c. Topicality hierarchy: [ht] > [+ht]
In terms of these feature specifications, topic-prominent languages can be saidto link [ht] (topically most salient) to subject; subject-prominent languages,on the other hand, link [hr] (highest argument) to subject. Within this system,we can say that subject is cross-linguistically the default topic because [ht]is assigned to the argument with the [hr] feature12 by default when contextidentifies none of the arguments as topically salient (or non-salient).
Returning to the Bantu agreement systems, we can characterize S-O re-
versal languages as having salience-based topic marking in the sense that the
agreement prefix encodes the most topical, but not necessarily the highest, ar-
12[hr] reads as there is no argument that is higher in the argument hierarchy, and is assigned
to the highest role.
175
7/28/2019 Zaspil43 Morimoto
16/27
Yukiko Morimoto
gument. We might represent this property of the topic agreement in a lexical
entry like that in (27). These functional schemata, familiar from LFG, specify
that there is some feature structure (f1) that contains a grammaticized discourse
topic TOP, whose value is another feature structure (f2). This inner feature
structure contains the information specified by the topic marker (TM), gender
and number. The optional semantic content SEM = pro abbreviates two fea-
ture structures, one with pro, which represents a TM functioning as a topic
pronoun, and the other without pro, representing a TM as topic agreement.
(27) Lexical entry of the topic marker (TM)
TM-: Vinfl (f1 TOP) = f2(f2 GEND) =
(f2 NUM) = ((f2 SEM) = pro)
The instantiations of these functional schemata result in the feature structure
shown in (28). The curved line from f2, the value ofTOP, to GF in (28) indicates
that the TOPIC is associated with one of the argument selected by the predicate
(realized as a core GF). This is ensured by the extended coherence conditionofLFG (Bresnan 2001, chapter 4), which requires that grammaticized discourse
functions TOPIC and FOCUS be associated with one of the argument functions
selected by the predicate.
(28)
f1
TO P f2
(SE M proi)
GEND
NU M
GF
V
TM -verb
Reversal lgs.
Non-reversal languages like Chiche wa employ role-based subject marking: the
subject marker (SM) encodes the highest argument in the argument hierarchy.
Thus, the lexical entry may contain the information shown in (29). Here the SM
specifies that there is some feature structure (f1) that contains a SUBJ, whose
value is identified with another feature structure (f2). The information includedin the inner f-structure is identical to that of the TM in (27). The f-structure in
(30) shows the instantiations of these functional schemata.
176
7/28/2019 Zaspil43 Morimoto
17/27
Agreement Properties and Word Order in Comparative Bantu
(29) Lexical entry of the subject marker (SM)
SM-: Vinfl (f1 SUBJ) = f2
(f2 GEND) = (f2 NUM) =
((f2 SEM) = pro)
(30)
f1
SUBJ f2
(PRED proi)
GEND
NU M
V
SM -verb
Chichewa
Now there is a third type, namely, languages such as Sesotho and Setswana
that do not allow S-O reversal like the salience-based agreement languages, but
also do not allow non-topical subjects to agree like the role-based agreement
languages (cf. Demuth and Mmusi 1997). In fact, I believe the majority of
Bantu languages are described in the literature as belonging to this type.
Within the proposed systems of agreement, we can characterize this third
type as having a basically role-based agreement system. The only restriction
is that the agreeing subject must be topically salient ([ht]). The f-structure in(31) shows that the TOPIC function is always associated with SUBJ.
(31)
f1
TO P f2
(SE M proi)
GEND
NU M
SUBJ
V
SM -verb
Sesotho/Setswana
From the diachronic perspective, the Sesotho/Setswana type can be viewed as
an intermediate stage in the shift from topic to subject agreement. One piece
of supporting data for the supposed historical path that gave rise to the two
agreement systems in the present day Bantu languages comes from Meeussens
(1967) reconstruction of Proto-Bantu. Meeussen (1967:120) reconstructs a syn-
tactic pattern in Proto-Bantu that is analogous to S-O reversal (and locative in-
version), as illustrated in (32)(33). These data lead us to infer that the system
of topic agreement already existed in Proto-Bantu.13
13The precise translation of (32) is not available in the original source.
177
7/28/2019 Zaspil43 Morimoto
18/27
Yukiko Morimoto
(32) a. nkma j-ji buenge bua-mit
S (exp) SM-V O (theme)
b. buenge bua-mit bu-ji nkmaO (theme) SM-V S (exp)
The monkey knows the (cleverness of) trees.
(33) a. njogu j-akuide mu-du
S (exp) SM-V LOC
The elephant died in the river.
b. mu-du mu-akuide njogu
LOC SM-V S (exp)In the river (there) died an elephant.
The historical path from topic to subject marking suggested by the above data
can be attributed to a frequentist explanation recently advocated by researchers
like Bybee (Bybee 1998, Bybee and Hopper 2001) and Haspelmath (2005) to
explain certain morphosyntactic regularities and change. The apparent change
in the agreement systems observed through the synchronic variation across the
Bantu family can be seen as an effect of frequencynamely, the notion of sub-ject embodies a frequent combination of semantic and information-structural
properties, most prominently a combination of agents and topics. Subject agree-
ment is more iconic than topic agreement in the sense that there is one-to-one
correspondence between the agreement morphology and grammatical function
(paradigmatic isomorphism, cf. Haiman 1980, Croft 1990a). These func-
tional forces seem to be reasonable motivations for the shift in which the topic
agreement was reanalyzed as subject agreement, and consequently the prever-
bal agreeing NP became restricted to the subject function. We can then also
expect a further historical path where all Bantu languages eventually come to
employ the role-based agreement system. The expected concomitant change
is, of course, the eventual loss of S-O reversal. Additional data on S-O reversal
suggest that this may already be happening, as discussed below.
4 Grammatical Change along Referential Hierarchies
When the new system replaces the old, it typically does so gradually in a pre-dictable fashion along referential hierarchies. A case in point is split ergativity.
Split ergativity is viewed as a transitional synchronic stage in a change from the
178
7/28/2019 Zaspil43 Morimoto
19/27
Agreement Properties and Word Order in Comparative Bantu
ergative/absolutive system to nominative/accusative system (cf. Dixon 1994).
In ergative languages, it is claimed that the higher the NP on the nominal hi-
erarchy in (34), adopted by Dixon (1994:85), the more likely it is to be in A
(transitive subject) rather than O (transitive object) function (e.g. Du Bois 1987,
Dixon 1994).
(34) Nominal Hierarchy
1st > 2nd > 3rd > Name > Human > Animate > Inanimate
more likely to be in A than in O function
The first person pronoun is therefore said to be the most prototypical or un-
marked A, while the inanimate NP is the most unmarked O. Assuming the
general markedness principle widely assumed that marked forms are also mor-phosyntactically marked, what we expect in a split ergative language along this
nominal hierarchy is presence of morphological marking of an NP from the
lowest element of the hierarchy when it is in A function, but from the highest
element in O function. According to Dixon (1994:85), a number of languages
have split case-marking systems exactly on this principle: a morphologically
overt (=marked) ergative case is used with NPs from the lowest end, up to
some point in the middle of the hierarchy, and an accusative case from that point
on, over to the highest end of the hierarchy. Split ergativity illustrates thus how
a shift from one system to another takes place systematically and predictably
along a referential hierarchy.
In the Bantu languages with the salience-based agreement system, we see
that S-O reversal is restricted along the dimension of the animacy hierarchy (= a
simpler version of the nominal hierarchy in (34)) given in (35).
(35) Animacy hierarchy: Human > Animate > Inanimate
As in the nominal hierarchy in (34), given two arguments, subject and ob-ject, the highest element in the hierarchy (human) is the most prototypi-
cal/unmarked, orto use the notion of frequency insteadmost frequently
the subject, while the lowest element (inanimate) is the most frequently the ob-
ject. Then presumably the most frequent structure (and animacy configuration)
would be one such as the boy is reading a book where the subject is human,
and the object, inanimate. On the other hand, the least frequent (animacy) con-
figuration would be one like loud noise disturbed the baby, where the subject is
inanimate, and the object, human.As reported by Morimoto (2003), it is these predicates with rare (termed
marked in Morimoto 2003) animacy relations that block S-O reversal. This
179
7/28/2019 Zaspil43 Morimoto
20/27
Yukiko Morimoto
is exemplified in (36)(37).14 Example (36) (presumably) represents the least
frequent animacy configuration, where the subject is inanimate and the object
is human. As shown in the translation for (36), the reversal reading is blocked.
The b sentence is grammatical only in the non-reversal, non-sensical reading.In (37), the subject is also lower in animacy, and again, the reversal reading is
blocked in the b sentence.
(36) a. Urushiinge ru-ra-joomb-a umwaana. S=inanimate
needle it-AF-pierce- AS P child O=human
The needle will pierce the child.
b. Umwaana a-joomb-a urushiinge. *reversal
child he-pierce-AS P needle
*The needleFOC will pierce the childTOP. Kinyarwanda
(37) a. Akayabu ka-a-ra-fyese umuhungu. S=animate
cat it-PAST-AF-lick:PERF boy O=human
The cat licked the boy.
b. Umuhungu a-a-fyese akayabu. *reversal
boy he-PAST-lick:PERF cat
*The catFOC licked the boyTOP. Kirundi
When the arguments are of equal animacy and there is potential ambiguity in
interpretation, as in (38), the reversal interpretation is unavailable just from the
string (in a null context); the only possible meaning in (38b) is one in which
umukoowa girl is the subject/agent, and umuhu ungu boy is the object/patient.
(38) a. Umuhuungu y-a-som-ye umukoowa. S=human
1boy 1-PS T-kiss-AS P 1girl O=human
The boy kissed the girl.
b. Umukoowa y-a-som-ye umuhuungu.
1girl 1-PS T-kiss-AS P 1boyThe girl kissed the boy. (no reversal)
*The boyFOC kissed the girlTOP. (reversal) Kinyarwanda
On the other hand reversal is allowed in an example like (39) where only one of
them can be the likely subject/agent due to the core meaning of the verb.
14The examples in (36)(39) are taken from Morimoto (2003).
180
7/28/2019 Zaspil43 Morimoto
21/27
Agreement Properties and Word Order in Comparative Bantu
(39) a. Icyuma cy-a-kas-e umugaati. S=inanimate
knife it-PS T-cut-AS P bread O=inanimate
The knife cut the bread.
b. Umugati w-a-kas-e icyuma.
reversal
bread it-PS T-cut-AS P knife
The knifeFOC cut the breadTOP. Kinyarwanda
In short, S-O reversal is permitted when the subject outranks the object in ani-
macy, but not when the subject is lower in animacy than the object. When there
is equal animacy and ambiguity results, the reversal interpretation becomes un-
available. These facts are summarized in (40).
(40) Grammatical Function-Animacy Association in S-O Reversal
OB J
HUM ANIM INAN
HU M yes* yes yes
SU BJ A NI M no yes* yes
INAN no no yes*
yes*: only if there is no ambiguity
The rationale for the facts summarized in (40) seems straightforward: on the one
hand, the configuration in which object outranks subject in animacy is less fre-
quent than the one in which subject outranks object. On the other hand, S-O re-
versal sentences are non-canonicalpresumably also rarer in frequencyboth
syntactically (non-canonical OVS order) and pragmatically (patient is topical
and agent is focal), than the non-reversal counterparts. The descriptive gener-
alization, then, is that the less frequently occurring animacy relations cannot be
expressed in the non-canonical syntactic construction (reversal).15
This type ofnon-canonical form-meaning pairs that are not likely to appear frequently in the
language (or blocked altogether as in the present case) seems to be susceptible
to change/loss over time (cf. Croft 1990b).16
In short, the animacy restriction on S-O reversal suggests this construction
is being lost from the least frequent combination of elements in the referen-
15See Morimoto (2001) for an OT analysis of these facts.16Animacy also figures in the domain of differential object marking (DOM; cf. Bossong
1985, Aissen 2003), in which objects that are high in animacy are case-marked, while those
low in animacy are unmarked. Like split ergativity, DOM is also said to represent a stage in the
diachronic process whereby overt case-marking is being lost from the low end of the referential
hierarchy.
181
7/28/2019 Zaspil43 Morimoto
22/27
Yukiko Morimoto
tial and argument hierarchy (e.g. inanimate-subject, human-object). And this
change must be concomitant to the change in the properties of the agreement
morphology. Properly recognizing the two systems in the Bantu family, the
salience-based and role-based, then enables us to relate the relevant synchronic
data like the animacy restriction on S-O reversal, inventory of non-topical sub-
jects to the issue of agreement and to a diachrony of the agreement systems.
5 Conclusion
In this final section I summarize the findings and main issues discussed in this
paper, and then point out some remaining issues for future work.
5.1 Summary
In this paper, I have argued for the two-fold agreement systems in Bantu:
salience-based agreement and role-based agreement. The crucial evidence for
salience-based agreement comes from the existence of S-O reversal, where the
preverbal patient triggers topic, rather than subject, agreement. The subject
analysis of this construction that assumes grammatical relation change was
rejected on both empirical and theoretical basis. I have shown that the pro-
posed view of agreement better explains the set of covariation related to the
presence/absence ofS-O reversal.
I have also tried to relate the synchronic data to the historical path from
topic to subject agreement, which I suggest is a particular realization of the
more general functional shift from topic to subject. Referential properties like
animacy seem to figure in diachronic change, as attested in split ergativity and
differential object marking. The animacy restriction on S-O reversal thus proba-
bly represents a stage in the gradual loss of this construction, concomitant with
the shift from topic to subject agreement.
In conclusion, in order to provide a coherent explanation of the set of facts
related to S-O reversal and to relate those facts to a broader set of agreement
phenomena (split ergativity, DOM), it is crucial that we recognize the role of
topicality in the core grammar, as manifested in topic agreement.
5.2 Remaining Issues
The present proposal that Bantu languages split between topic and subjectagreement has naturally opened doors to new issues and problems, some of
which are clearly beyond the scope of this paper. Below I address a few note-
182
7/28/2019 Zaspil43 Morimoto
23/27
Agreement Properties and Word Order in Comparative Bantu
worthy issues for future work.17
First, given the claim that a subset of Bantu languages display topic agree-
ment, we should expect more differences between topic agreement and sub-
ject agreement languages than those discussed in section 2. For example, Kin-yarwanda has four pleonastic pronouns (only for subjects) that are used in im-
personal constructions. Under the topic agreement analysis, one would need ad-
ditional explanation of how such impersonal pronouns can exist at all. Kimenyi
(1980:185) notes that one of these pleonastic pronouns that refers to an unspec-
ified agent, ba- (they), always has ambiguous interpretation as a dummy
pronoun and a definite personal pronoun. There are also cases where this pro-
noun must be interpreted as a definite, topic-anaphoric pronoun. It would be
worthwhile to examine the properties of these impersonal pronouns in both
topic agreement and subject agreement languages.
Second, from the historical perspective, it is interesting to note that while
S-O reversal is restricted to only a subset of Bantu languages, locative inversion
remains widespread throughout this language family. As we have seen, S-O
reversal generally involves inversion of a proto-typical agent and patient. The
animacy condition we saw in section 4 limits S-O reversal to only a subset of
predicates. Such a split within the construction makes the overall frequency
of S-O reversal relative low compared to a construction without such a split
(due to the animacy restriction). It might well be that the decrease in overallfrequency has led to the total loss of this construction in the majority of the
Bantu languages. To pursue this view, one would have to examine quantitative
data on S-O reversal as well as locative inversion.
Lastly, in order to establish a more solid argument for the existence of
topic agreement, more contextualized data (e.g. from text) would perhaps be
necessary.
Nonetheless, I hope that the present discussion on the split between sub-
ject and topic agreement across the Bantu family serves as a step forward to-wards a better understanding of the relation between agreement properties and
word order in Bantu.
6 References
Aissen, J. (2003). Differential case marking: Iconicity vs. economy. Natural
Language & Linguistic Theory, 21:435483.
Alsagoff, L. (1992). Topic in Malay: The Other Subject. Ph.D. thesis, StanfordUniversity, Stanford, CA.17The remaining issues addressed here have been prompted by the editors of this volume.
183
7/28/2019 Zaspil43 Morimoto
24/27
Yukiko Morimoto
Birner, B.J. & Ward G. (1998). Information Status and Noncanonical
Word Order in English. Studies in Language Companion Series. Amster-
dam/Philadelphia: John Benjamins.
Bokamba, E.G. (1976). Question Formation in Some Bantu Languages. Ph.D.thesis, Indiana University, Indiana.
Bokamba, E.G. (1979). Inversions as grammatical relation changing rules in
Bantu. Studies in the Linguistic Sciences, 9:124.
Bokamba, E.G. (1981). Aspects of Bantu Syntax. Preliminary edition. Depart-
ment of Linguistics, University of Illinois, Urbana-Champaign, Illinois.
Bokamba, E.G. (1985). Verbal agreement as a noncyclic rule in Bantu. In: D.L.
Goyvaerts (ed.) African Linguistics: Essays in memory of M.W.K Semikenke,
954. Amsterdam/Philadelphia: John Benjamins.
Bossong, G. (1985). Differentielle Objektmarkierung in den neuiranischen
Sprachen. Tubingen: Gunter Narr Verlag.
Bresnan, J. (ed.) (1982). The Mental Representation of Grammatical Relations.
Cambridge, MA: The MIT Press.
Bresnan, J. (1994). Locative inversion and universal grammar. Language,70:72131.
Bresnan, J. (2001). Lexical-Functional Syntax. Oxford: Blackwell Publishers.
Bresnan, J. & Kanerva J.M. (1989). Locative inversion in Chichewa: A case
study of factorization of grammar. Linguistic Inquiry, 20(1):150.
Bresnan, J. & Mchombo S.A. (1987). Topic, pronoun, and agreement in
Chichewa. Language, 63:741782.
Bybee, J. (1998). A functionalist approach to grammar and its evolution. Evo-
lution of Communication, 2:249278.
Bybee, J. & Hopper, P. (2001). Frequency and the emergence of linguistic
structure. Amsterdam: John Benjamins.
Croft, W. (1990a). Possible verbs and the structure of events. In: S. Tsohatzidis
(ed.) Meanings and prototypes: Studies in linguistic categorization, 4873.
London: Routledge.
Croft, W. (1990b). Typology and Universals. Cambridge Textbooks in Linguis-
tics. Cambridge: Cambridge University Press.
184
7/28/2019 Zaspil43 Morimoto
25/27
Agreement Properties and Word Order in Comparative Bantu
Dalrymple, M., Kaplan, R.M., Maxwell, J.T., & Zaenen A. (eds.) (1995). For-
mal Issues in Lexical-Functional Grammar. Stanford, CA: CSLI Publica-
tions.
Demuth, K. & Mmusi, S. (1997). Presentational focus and thematic structure incomparative Bantu. Journal of African Languages and Linguistics, 18:119.
Dixon, R.M. (1994). Ergativity. Cambridge, MA: Cambridge University Press.
Du Bois, J.W. (1987). The discourse basis of ergativity. Language, 63:805855.
Falk, Y.N. (1983). Subjects and long-distance dependencies. Linguistic analy-
sis, 12:245270.
Gazdar, G. (1981). Unbounded dependencies and coordinate structure. Linguis-
tic Inquiry, 12:155184.
Givon, T. (1976). Topic, pronoun, and grammatical agreement. In Li (1976),
149188.
Haiman, J. (1980). The iconicity of grammar. Language, 56:515540.
Harford, C. (1990). Locative inversion in Chishona. In: J.P. Hutchison &
V. Manfredi (eds.) Current Approaches to African Linguistics, volume 11,
137144. Dordrecht: Foris Publications.
Haspelmath, M. (2005). Against markedness (and what to replace it with). MS .
Max-Planck-Institut fur evolutionare Anthropologie.
Joshi, S. (1993). Selection of Subjects and Objects in Marathi. Ph.D. thesis,
Stanford University, Stanford, CA.
Keenan, E. (1976). Towards a universal definition of subject. In Li (1976),
303333.
Kimenyi, A. (1980).A Relational Grammar of Kinyarwanda
. Berkeley, CA:University of California Press.
Kimenyi, A. (1988). Passives in Kinyarwanda. In: M. Shibatani (ed.) Passive
and Voice, 355386. Amsterdam/Philadelphia: John Benjamins.
Kinyalolo, K.K.W. (1991). The SPEC-Head Agreement Hypothesis in KiLega.
Ph.D. thesis, University of California Los Angeles, Los Angeles, CA.
Kroeger, P. (1993). Phrase Structure and Grammatical Relations in Tagalog.
Dissertations in Linguistics. Stanford, CA: CSLI Publications.
Lambrecht, K. (1994). Information Structure and Sentence Form. Cambridge,
UK: Cambridge University Press.
185
7/28/2019 Zaspil43 Morimoto
26/27
Yukiko Morimoto
Li, C. (ed.) (1976). Subject and Topic. New York: Academic Press.
Li, C.N. & Thompson, S.A. (1976). Subject and topic: A new typology of
language. In: C. Li (ed.) Subject and Topic, 457489. New York: Academic
Press.
Ldrup, H. (1999). Linking and Optimality in the Norwegian presentational
focus construction. Nordic Journal of Linguistics, 22:205230.
Machobane, M.M. (1995). The Sesotho locative constructions. Journal of
African Languages and Linguistics, 16:115136.
Manning, C.D. (1996). Ergativity: Argument Structure and Grammatical Rela-
tions. Dissertations in Linguistics. Stanford, CA: CSLI Publications.
Meeussen, A. (1967). Bantu grammatical reconstructions. Africana Linguistica
III, 61:79121. Tervuren Belgique Royal Museee de Lafrique: Sciences
Humaines Annales.
Morimoto, Y. (1999). An Optimality account of argument reversal. In: M. Butt
& T.H. King (eds.) Proceedings of the LFG 99 Conference. Stanford, CA:
CSLI Publications. http://csli-publications.stanford.edu/
LFG/4/lfg99.html.
Morimoto, Y. (2000). Discourse Configurationality in Bantu Morphosyntax.Ph.D. thesis, Stanford University, Stanford, CA.
Morimoto, Y. (2002). Prominence mismatches and differential object
marking in Bantu. In: M. Butt & T.H. King (eds.) Proceedings of
the LFG-02 Conference. Stanford, CA: CSLI Publications. http://
csli-publications.stanford.edu/LFG/6/lfg02.html .
Morimoto, Y. (2003). Markedness hierarchies and optimality in Bantu. In:
R. Laury, G. McMenamin, S. Okamoto, V. Samiian, & K. Subbarao (eds.)
Perspectives in Linguistics: Papers in Honor of P.J Mistry, 261280. New
Delhi: Indian Institute of Language Studies.
Ndayiragije, J. (1999). Checking economy. Linguistic Inquiry, 30:399444.
Stucky, S. (1981). Word Order Variation in Makua: A Phrase Structure Gram-
mar Analysis. Ph.D. thesis, University of Illinois, Urbana, IL.
Stucky, S. (1983). Verb phrase constituency and linear order in Makua. In:
G. Gazdar (ed.) Order, Concord, and Constituency, 7594. Dordrecht: Foris.
Ura, H. (1996). Multiple Feature Checking. Ph.D. thesis, MIT, Cambridge,
MA.
186
7/28/2019 Zaspil43 Morimoto
27/27
Agreement Properties and Word Order in Comparative Bantu
Wald, B. (1979). The development of the Swahili object marker: A study of the
interaction of syntax and discourse. In: T. Givon (ed.) Discourse and Syntax,
volume 12 ofSyntax and Semantics, 505524. New York: Academic Press.
Williams, E.S. (1977). Across-the-board application of rules. Linguistic In-quiry, 8:419423.
Woolford, E. (1987). An ECP account of constraints on across-the-board ex-
traction. Linguistic Inquiry, 18:166171.
Wunderlich, D. (2000). Predicate composition and argument extension as gen-
eral optionsa study in the interface of semantic and conceptual structure.
In: B. Stiebels & D. Wunderlich (eds.) Lexicon in Focus, 249272. Berlin:
Akademie Verlag.